eightforge-performance-reviewer

Invoke the performance reviewer on the changed files and it will classify each path, check the non-negotiable rules, and walk the review checklist. It returns a Pass, Pass with Concerns, or Fail verdict with severity-ordered issues and minimal fixes.

It flags unbounded row reads, missing pagination, unselective filters, and hot paths where RLS predicate cost has no measured query plan. Deep RLS policy correctness itself is routed to the dedicated Supabase reviewer.

Yes, it evaluates Server Components versus client hydration choices for data-heavy surfaces and checks that React lists avoid full re-renders of large structures. It also verifies API routes bound work with timeouts and batch sizes.

Lossy optimizations include silent row truncation, silent sampling, dropped evidence dependencies, weakened tenant isolation, bypassed freshness, and replacing durable state with an unauditable cache. Any of these causes a Fail verdict regardless of speed gains.

Route extraction completeness to the document intelligence reviewer, RLS and query-policy correctness to the Supabase reviewer, truth and freshness authority to the truth engine reviewer, and interaction usability to the UX reviewer. This reviewer owns correctness-preserving scale only.